基于选择性集成SVM的数据类型识别

提出基于选择性集成支持向量机的语音、话带数据信号分类方法，根据集成算法的差异性定义，采用两层级联结构的动态叠加算法完成决策输出。该方法能够在训练阶段准确地选择具有较高识别精度和差异性的成员分类器，在测试阶段对各成员分类器进行动态集成，保证最终的分类结果最优。构建时域、频域相结合的特征向量，并具有较好的抗噪声能力。实验结果表明，该方法无论在分类还是在运算复杂度上都取得较好的效果。

Data Type Recognition Based on Selective Ensemble SVM

This paper proposes a voice and Voice Band Data（VBD） classification method based on selective ensemble Support Vector Machine（SVM）.According to a difference definition of ensemble algorithm,the classifier output is completed by two level structure’s dynamic stacking algorithm.In the training stage,the method can choose accurately the number classifiers which have the high recognition precision and the difference.In the test stage,a dynamic ensemble algorithm is used to guarantee that the final classified result is most superior.In the characteristic parameters extraction,the character vector is built that includes the time domain and the frequency domain character.These parameters have good anti-noise ability.Experimental results show that the algorithm that is proposed in the method has good classification effect and low operation complexity.